基于增强CT的瘤内-瘤周影像组学诊断肾嗜酸细胞腺瘤和嫌色细胞癌:一项多中心回顾性分析  

Intratumoral and peritumoral Radiomics based on multi-phase contrast-enhanced CT for diagnosis of renal oncocytoma and chromophobe renal cell carcinoma:a multicenter retrospective study

作　　者：林文香 冯文辉 叶泳松[3] 翁蓓 谢尚煌[5] 钟桂棉 刘天柱[7] LIN Wenxiang;FENG Wenhui;YE Yongsong;WENG Bei;XIE Shanghuang;ZHONG Guimian;LIUTianzhu(Department of Imaging,Hengqin Hospital,the First Affiliated Hospital of Guangzhou Medical University(Hengqin Central Hospital of Guangdong-Macao Deep Cooperation Zone),Zhuhai,Guangdong,China,519000;Imaging Department of Zhuhai People's Hospital,Zhuhai,Guangdong,China,519099;Imaging Department of Ersha Branch of Guangdong Hospital of Traditional Chinese Medicine,Guangzhou,Guangdong,China,510120;Department of Imaging,The First Affiliated Hospital of Sun Yat-sen University,Guangzhou,Guangdong,China,510080;Department of Medical Imaging,Longgang Central Hospital,Shenzhen Guangdong,China,518116;Imaging Department,Panyu Central Hospital Affiliated to Guangzhou Medical University,Guangzhou,Guangdong,China,511431;Imaging Department,Zhuhai Hospital,Guangdong Hospital of Traditional Chinese Medicine,Zhuhai Guangdong,China,519015)

机构地区：[1]广州医科大学附属第一医院横琴医院(横琴粤澳深度合作区中心医院)影像科,广东珠海519000 [2]珠海市人民医院影像科,广东珠海519099 [3]广东省中医院二沙分院影像科,广东广州510120 [4]中山大学附属第一医院影像科,广东广州510080 [5]深圳市龙岗中心医院医学影像科,广东深圳518116 [6]广州医科大学附属番禺中心医院影像科,广东广州511431 [7]广东省中医院珠海医院影像科,广东珠海519015

出　　处：《分子诊断与治疗杂志》2025年第2期230-234,共5页Journal of Molecular Diagnostics and Therapy

基　　金：珠海市卫生健康局项目(2220009000170)。

摘　　要：目的本研究旨在结合皮质髓质期(CMP)和肾实质期(NP)的瘤内及瘤周影像组学特征构建肾嗜酸细胞腺瘤(RO)和嫌色细胞癌(CRCC)的诊断模型,并将相关模型与人工诊断结果进行比较。方法将RO和CRCC病例分为训练组(70%)和验证组(30%)。提取CMP和NP瘤内和瘤周(3 mm)的CT影像组学特征。图像分割由放射科医生手动执行,瘤周掩膜由Python执行。采用单变量分析、最小绝对收缩和选择算子及组内相关系数对获得的影像组学特征进行筛选,并通过机器学习的随机森林(RF)算法完成建模。另外所有图像由两名放射科医生手动诊断并记录诊断结果。使用校准、决策和ROC曲线对RF模型进行评估。比较RF模型与手工诊断的准确性、敏感性和特异性。结果92例患者中,有65例(29例RO,36例CRCC)在训练集,27例(12例RO,15例CRCC)在验证集。训练集中各个模型均表现优异。验证集中,最佳验证模型为模型2和模型5[曲线下面积(AUC)分别为0.825及0.872],最差验证模型为模型3(AUC=0.639)。CMP模型(模型1和模型2)总体优于NP模型(模型3和模型4)。RF模型(灵敏度:60.00~88.89%;特异性:73.33~81.25%;准确率:66.67~81.48%)优于人工诊断(灵敏度:46.74~70.59%;特异性:41.67~46.34%;准确率:52.27~59.78%)。结论CMP瘤内和瘤周放射组学模型对RO和CRCC的诊断有价值且优于人工诊断。Objective This study aimed to construct diagnostic models for renal oncocytoma(RO)and chromophobe renal cell carcinoma(CRCC)using corticomedullary phase(CMP)and nephrographic phase(NP)intratumoral and peritumoral radiomics features,and to compare these models with manual results.Methods RO and CRCC cases from five centers were split into training(70%)and validation(30%)sets.CMP and NP intratumoral and peritumoral 3 mm CT radiomic features were extracted.Segmentation was performed by radiologists and Python.Features were selected using univariate analysis and least absolute shrinkage,selection operator and intraclass correlation coefficients for Random Forest(RF)models.All images were assessed by two radiologists,and radiological reports were collected.RF models were evaluated using calibration,decision,and ROC curves.The accuracy,sensitivity,and specificity of models and manual diagnoses were compared.Results Of the 92 cases,65(29 RO,36 CRCC)were in the training set and 27(12 RO,15 CRCC)in the validation set.Training models showed excellent performance.The best validation model was model 2 and 5[area of under curve(AUC)=0.825 and 0.872],and the worst was model 3(AUC=0.639).CMP models(model 1 and 2)generally outperformed NP models(model 3 and 4).RF models(sensitivity:60.00~88.89%;specificity:73.33~81.25%;accuracy:66.67~81.48%)outperformed manual diagnosis(sensitivity:46.74~70.59%;specificity:41.67~46.34%;accuracy:52.27~59.78%).Conclusion CMP intratumoral and peritumoral radiomics models are valuable for diagnosing RO and CRCC and better than manual diagnosis.

关 键 词：嫌色细胞癌 嗜酸细胞腺瘤 影像组学 计算机横断扫描 

分 类 号：F20[经济管理—国民经济]